What Matters to Users? Factors that Affect Users’ Willingness to Share Information with Online Advertisers Pedro Giovanni Leon∗, Blase Ur∗, Yang Wangz, Manya Sleeper∗, Rebecca Balebako∗, Richard Shay∗, Lujo Bauer∗, Mihai Christodorescuy, Lorrie Faith Cranor∗ ∗Carnegie Mellon University yQualcomm Research Silicon Valley zSyracuse University {pedrogln, bur, msleeper, balebako, [email protected] [email protected] rshay, lbauer, lorrie}@cmu.edu ABSTRACT 1. INTRODUCTION Much of the debate surrounding online behavioral adver- Online behavioral advertising (OBA), the practice of tar- tising (OBA) has centered on how to provide users with geting online advertising based on users' past online activi- notice and choice. An important element left unexplored ties, has been the subject of a major privacy debate in recent is how advertising companies' privacy practices affect users' years. Reports released in 2012 by the U.S. Federal Trade attitudes toward data sharing. We present the results of Commission [5] and the White House [36] discuss the pri- a 2,912-participant online study investigating how facets of vacy tradeoffs inherent in this practice. At the same time, privacy practices|data retention, access to collected data, browser vendors have recently taken steps to reduce track- and scope of use|affect users' willingness to allow the col- ing: Microsoft sends a Do Not Track signal by default in IE lection of behavioral data. We asked participants to visit a 10 [22], and Mozilla has announced that Firefox will even- health website, explained OBA to them, and outlined poli- tually block third-party cookies by default [6]. cies governing data collection for OBA purposes. These poli- As battles rage about default behaviors and options, av- cies varied by condition. We then asked participants about erage users are asked to make choices about their privacy their willingness to permit the collection of 30 types of infor- preferences regarding online behavioral advertising. In some mation. We identified classes of information that most par- cases, these choices have limited granularity. For instance, ticipants would not share, as well as classes that nearly half with the Do Not Track signal under debate [31], users have of participants would share. More restrictive data-retention the choice of actively turning Do Not Track on or off, or and scope-of-use policies increased participants' willingness leaving it unset. In many other cases, however, users have to allow data collection. In contrast, whether the data was a more complex decision to make. As part of the advertis- collected on a well-known site and whether users could re- ing industry's self-regulation program, users can opt out of view and modify their data had minimal impact. We discuss behavioral advertising from individual companies [25]. Simi- public policy implications and improvements to user inter- larly, third-party privacy tools like Abine's DoNotTrackMe1 faces to align with users' privacy preferences. and Evidon's Ghostery2 enable users to see which compa- nies are tracking their activities on a particular site, and to Categories and Subject Descriptors block particular companies. In past work, researchers have found that familiarity with a third-party tracking company K.4.1 [Computers and Society]: Public Policy Issues| influences users' attitudes about data collection [34]. Privacy Unfortunately, little is known about other factors that may influence users' preferences. For instance, does the General Terms length of time behavioral data is retained actually matter to most users? Does it make a difference whether data is Human Factors, Design used to target advertisements only on a single first-party website, or on Facebook, or on any website on the Inter- Keywords net? This understanding is crucial for the design of future Privacy, User Preferences, Online Behavioral Advertising, OBA privacy tools. For instance, when a privacy tool asks OBA, Tracking, Data Retention, Do Not Track the user to decide whether to permit or block the collection of data by a particular entity, the tool could highlight that entity's privacy practices that most strongly affect users' de- cisions. Better understanding the drivers of user behavior might also influence public policy. For instance, laws and regulations designed to support consumer privacy could fo- cus on practices that most affect users' comfort with data collection and sharing, rather than focusing on distinctions Copyright is held by the author/owner. Permission to make digital or hard that have little bearing on users' preferences. copies of all or part of this work for personal or classroom use is granted without fee. 1 https://www.abine.com Symposium on Usable Privacy and Security (SOUPS) 2013, July 24–26, 2 2013, Newcastle, UK. http://www.ghostery.com 1 In this paper, we examine how four dimensions of privacy allowing users to be linked to one or more social networking practices impact users' willingness to permit the collection profiles. They also found that some websites directly leaked of data for OBA. These dimensions are the length of time personally identifiable information [15]. Roosendaal found data will be retained, whether or not a user will have access that Facebook tracked both users and non-users of Facebook to review and modify this data, the range of websites on across the Internet using cookies attached to \Like" buttons which advertising will be targeted based on this data, and embedded in other pages [28]. whether the data was collected on a well-known website. This large data footprint leads to privacy concerns. Re- To this end, we conducted a 2,912-participant online sur- tailers can combine credit or debit card histories with data vey. We asked participants to visit a health website. After from online tracking to create detailed customer profiles re- they explored this page, we explained the value proposi- vealing potentially sensitive \lifestyle or medical issue[s]" [9]. tion of online behavioral advertising: that advertising and Even when data is collected in an aggregated, ostensibly the collection of data for targeted ads enable websites to be anonymized manner, bulk collection leaves the potential for free. We then showed the participant this website's data- re-identification [7, 24]. collection practices, with details varied based on the partic- Users tend to dislike the idea of being tracked and profiled ipant's condition. In different conditions, participants were by third parties [8]. Based on 48 semi-structured interviews, told that data would be retained for one day or indefinitely; Ur et al. found that participants recognized both pros and they were told or not told that they would be provided ac- cons of OBA, pointing out that OBA can be useful to both cess to review and modify collected data; participants were users and companies, yet also privacy-invasive. Participants told that data would be used for targeted advertising only did not understand how ads were targeted and believed com- on the health site, on both the health site and Facebook, panies collected more information than they generally col- or on any website; and the health site itself was either well lect [34]. Respondents to a 2011 survey conducted by Mc- known or a site we invented. We then asked participants Donald and Peha also believed that websites could collect to rate their willingness to allow the collection of 30 differ- more data than is currently accessible [21]. Wills and Zeli- ent types of information, and to answer additional questions jkovic created a JavaScript tool to show a user's location, related to their OBA preferences. age range and gender, all of which could be determined by Nearly half of our participants were unwilling to allow the third-party sites. Approximately half of their 1,800 partici- collection of any data, while the site's privacy practices im- pants were concerned about third-party tracking, the level of pacted the remaining participants' attitudes. Of the four data collection, and the ability of third-party data trackers dimensions we examined, the scope of use and the period of to infer demographic information after seeing these data [38]. data retention had the greatest impact on participants' will- OBA is currently self-regulated under guidelines created ingness to allow their information to be collected. Having by the advertising industry. These guidelines require that access to view and modify data collected, as well as par- users be provided the option to opt out of targeted ads [25], ticipants' familiarity with the website on which data was but do not mandate options for fine-grained control [19]. Un- being collected, did not appear to affect their willingness to der current guidelines, users are notified of the presence of allow data collection, at least in the narrow scenario we in- behavioral advertising through an AdChoices icon that ap- vestigated. Furthermore, the majority of participants were pears near or inside the ad. However, Leon et al. found that not willing to pay any money to prevent data collection or most users did not understand the purpose of this icon [17]. remove advertising, believing websites should be free. Users are currently able to limit online behavioral adver- We provide background on the debate surrounding online tising in several ways. By clicking on the AdChoices icon, behavioral advertising and highlight related work in Sec- users can visit an opt-out page that allows them to opt out of tion2. In Section3, we describe our methodology. We ad networks individually. Alternatively, they can use third- present our results in Section4, beginning with our factor party tools that block tracking, or even block ads entirely. analysis and proceeding through the four dimensions of pri- However, there are problems with all of these opt-out mech- vacy practices we investigated. We discuss our results in anisms. For instance, Komanduri et al. found many gaps Section5 before concluding in Section6.